wide 1
Semiconductor Expert Forum

Posted on

Cadence Claims the CFD High Ground with a New GPU-Based Accelerator

Cadence Claims the CFD High Ground with a New GPU-Based Accelerator
by Bernard Murphy on 02-01-2024 at 6:00 am
Categories: AI, Automotive, Cadence, EDA

J24135 Millennium Press Image 400x400 min

For observers of EDA markets there is an easily overlooked opportunity for new growth. Today around 50% of EDA revenues come from systems rather than semiconductor companies, from datacenters to automotive, aerospace, energy, and others. In most of these industries total system design depends as much on mechanical and other multiphysics optimizations (aerodynamics, stress, thermal, electromagnetics, etc.) as on electronic design. Multiphysics analysis has already penetrated semiconductor design, for example in-package to in-system thermal analysis and management using computational fluid dynamics (CFD) for cooling analytics. In short, multiphysics bridges between electronic system design and total system design as critical to support power generation, airline, and automotive markets. As in chip design, system problems in these domains keep getting harder, demanding active and continuous innovation from solution providers to address modern design needs. Exploiting synergies between EDA and multiphysics expertise, Cadence claims the Millennium platform delivers a breathtaking performance advance for multiphysics analytics, solving industrial scale problems in hours rather than weeks and opening large new growth opportunities.

A Fast Layman’s Guide to CFD

CFD simulates the flow of a fluid (liquid or gas) around/through mechanical structures like circuit boards, aircraft, gas turbines and cars. Without CFD, these measurements must be made on prototypes, for example in wind tunnels, an expensive and time-consuming process. With CFD, engineers can shift-left (a familiar concept in EDA), to study performance of a digital twin against simulated fluid flows.

Simulations are based on the Navier-Stokes differential equation, mapped across discrete meshes to enable numerical solving. Meshes are designed for finer spacing around critical zones with coarser spacing elsewhere and commonly run to many millions of elements. Factors considered in solving across the mesh include pressure and temperature, also viscosity because all fluids flow slower near boundaries. Compressibility can be important when considering acoustics or Mach speeds; turbulence is another factor at high speeds. These factors have enough impact on mesh and solver methods that CFD must provide a family of technology solutions.

Turbulence is the most challenging condition to simulate accurately. The most widely practiced technique in industry today develops static averages, a weak approximation for a dynamic phenomenon able to deliver accurate CFD around an airplane wing at cruising altitude but not during ascent or descent. A different technique called Large Eddy Simulation (LES) can model much more accurately and dynamically but is more expensive in computation, making extensive turbulence modeling through a digital twin impractical. Thus critical analyses have been limited to real physical modeling using prototypes in wind tunnels, effective but too cumbersome to explore thousands of scenarios for optimization.

Cadence Authority in CFD and LES

CFD is a high expertise domain with a lot of history. Tool departments and often product teams are staffed with armies of PhDs. Algorithms for meshes and solvers, together with software, have evolved significantly and of course continue to evolve. In other words this is a domain an EDA company must enter inorganically.

Cadence started here in 2021 with a series of acquisitions. These include NUMECA with strong meshing and solver technologies and an established reputation in marine and turbomachinery applications. Shortly after Cadence acquired Pointwise with proven strength in CFD meshing and established in aerospace and defense markets. In late 2022 they acquired Cascade Technologies, a Stanford spinout with compelling technology for LES. Through these acquisitions Cadence has built a stable of thoroughbred technology and experts in CFD, adding to their established strength in other aspects of multiphysics. But it seems they didn’t stop there.

Industries are desperate for higher LES performance for more accurate digital twin modeling. As one example, 50% of the energy consumed by a car goes to overcoming aerodynamic drag, directly affecting ICE fuel consumption or EV range. Designers need digital twins to simulate over thousands of operating conditions to find and optimize the many small improvements they can make around the car structure to reduce drag. How did Cadence step up to this need?

Cadence Millennium M1 and the Fidelity LES Solver

CFD is very parallelizable, so an obvious solution is run a job across many server/CPU clusters. This was already possible on big CPU farms or supercomputers, but cost becomes prohibitive when running complex LES algorithms over very large meshes with experiments over thousand of runs. Overcoming this barrier has been one of the drivers prompting development of Millennium M1, Cadence’s first GPU-based accelerator.

Cadence has a proven track record in hardware acceleration across multiple generations of the Palladium and Protium platforms for hardware verification. They have worked out the design, operations, and supply chain kinks to build these platforms and they have established infrastructure to provide cloud-based access. (All platforms including Millennium can also be purchased for on-premises analysis.) Extending this expertise to a GPU-based platform is both obvious and brilliant. In one stroke (though I’m sure it took them time to get there 😀) they can accelerate CFD simulations. Adding new generative AI methods for design and analysis exploration they claim delivers up to 100X design impact in accuracy, speed, and scale at much lower power when compared with massive CPU server parallelism. Hardware acceleration from Cadence hardware know-how combined with genAI expertise from both EDA and CFD teams demonstrates the synergy required to deliver the active and continuous innovation I mentioned earlier.

CFD algorithm development has also been very active. Software is designed from the ground up to be GPU native. Problem preparation for analysis includes low-touch optimized mesh generation. And there are new numerical methods to ensure high stability in LES simulations (normally prone to unphysical behavior in turbulence modeling).

This capability is available today for CFD multiphysics modeling, in the cloud or on-premises.

Millennium is Not Just for CFD

It is obvious that a GPU-based accelerator should be able to do more than accelerate CFD. It could accelerate finite element analyses such as stress, thermal diffusion, and electromagnetics. It can also run generative AI. But why not just use one of the giant hyperscaler GPU banks for that purpose? For me one reason is simply availability and latency in competition with chatbots and creative image apps. Equally it is hard to believe that application-specific fine-tuning on top of a mass market LLM models could serve the high complexity, high accuracy, and domain-specific needs of modern EDA and multiphysics software. Dedicated hardware is the way to go, accessible through the cloud or in on-premises installations.

It will be very interesting to see what capabilities Millennium will offer in the future both for electronic design and for multiphysics. You can learn more HERE.

Also Read:

2023 Retrospective. Innovation in Verification

Information Flow Tracking at RTL. Innovation in Verification

ML-Guided Model Abstraction. Innovation in Verification

Posted on

2024 Outlook with Justin Endo of Mixel

2024 Outlook with Justin Endo of Mixel
by Daniel Nenni on 01-31-2024 at 10:00 am
Categories: IP, Mixel

MixelLogo 25

 

Mixel is a semiconductor intellectual property (IP) company that we have been working with for 4 years with outstanding results.  Mixel’s focus is on developing mixed-signal IP, which includes analog and digital components. These IP cores are often used in applications such as mobile devices, consumer electronics, automotive systems, and other embedded systems where both analog and digital functions are required.

The company provides a wide range of IP cores for various interfaces and standards, including MIPI cores (Mobile Industry Processor Interface),  Multi-standard IP Cores,  and LVDS (Serializer and Deserializer). Their IP solutions aim to enable efficient and reliable communication between different components within electronic devices.

Tell us a little bit about yourself and your company.
Mixel is a leading provider of mixed-signal interface IPs. We offer a wide portfolio of high-performance mixed-signal connectivity solutions. Mixel’s mixed-signal portfolio includes PHYs and SerDes, such as MIPI PHYs (MIPI D-PHY, MIPI C-PHY, and MIPI M-PHY), LVDS, and Multi-standard SerDes. I lead Mixel’s marketing and sales strategy and implementation.

What was the most exciting high point of 2023 for your company?
2023 was a big year for us. We celebrated our 25th anniversary as a company. We hosted many of our partners, customers, and employees past and present at our global HQ in San Jose. It was great to reconnect with familiar faces and celebrate such a significant milestone with those who made it possible.

What was the biggest challenge your company faced in 2023?
Like many in our industry, there was a noticeable slowdown in the first half of 2023. While we are somewhat insulated as an IP provider, we still saw how dramatically things can change in a short period of time. Many of our customers pushed off their decision-making to later in the year. Thankfully, unlike many companies, small and large, that went through reduction of force, we increased our overall headcount. Last year was a good time to hire high quality talent due to the slow down, and a great time to focus on new product development.

How is your company’s work addressing this biggest challenge?
The best decision we made was to not slow down IP development during the downturn. After a slow first half of 2023, the second half was stronger than we expected. We are hoping to leverage this momentum coming into 2024 and we believe it will pay off in the coming months and years.

What do you think the biggest growth area for 2024 will be, and why?
As a leading provider of MIPI PHY IP, we have seen major growth in MIPI in automotive and AI enabled applications. As the number of sensors, cameras, and displays increase in cars, we see a corresponding increase in demand for automotive ICs on both sensors/display-side and processor-side. MIPI is seen as the de facto standard in many of those applications, particularly MIPI CSI-2 for low-power cameras and sensor.

How is your company’s work addressing this growth?
To meet our automotive customer’s requirements, we are ISO 26262 and ISO 9001 certified. Our process is certified up to ASIL-D and we have multiple IP configurations certified up to ASIL-B. We continue to develop for the latest MIPI specifications and having our ASIL-D certified process already in place allows us to provide automotive-grade IPs with functional safety documentation to support our customer’s safety critical applications.

What conferences did you attend in 2023 and how was the traffic?
As an active contributor to the MIPI Alliance and the development of its specifications, we attend all three face-to-face meetings each year. In addition, we sponsor many industry events including foundry exhibitions such as TSMC Technology Symposium & TSMC OIP Ecosystem Forum, GlobalFoundries Technology Summit, Samsung Foundry Forum & SAFE Forum, and Tower Semiconductor Technical Global Symposium At these events, we showcase our customers who have integrated our IP into their product, such as Microsoft’s HoloLens 2 and NXP’s i.MX7ULP applications processor. Attendees at these events really enjoy interacting with the technology and in terms of traffic to our booth, 2023 was the highest in recent years for several events, even surpassing pre-COVID numbers.

Will you attend conferences in 2024? Same or more?
We are aiming to attend the same events as last year and are actively looking to add others, particularly those outside the US, such as in China and Europe.

Additional questions or final comments?
Hope to see you at our next event!

Automotive-grade MIPI PHY IP drives multi-sensor solutions

MIPI D-PHY IP brings images on-chip for AI inference

MIPI bridging DSI-2 and CSI-2 Interfaces with an FPGA

Posted on

Chiplets Open Pandora’s Box

Chiplets Open Pandora’s Box
by Daniel Nenni on 01-31-2024 at 6:00 am
Categories: Chiplet, IP, Mirabilis Design

Chiplet

Chiplets have simplified one area of design but opened pandora’s box on another front. The simulation complexity of each chiplet is lower but now the chiplet-to-chiplet interconnect has become complex. Folks are experimenting with different interconnect protocols, variations of UCIe, modifying UCIe settings, interface speeds, number of physical layers and so one. Now add legacy standards like AXI, new protocols like PICe6.0 and cache coherency to the mix.

All-in-all, this creates a completely new set of experiments. One for which the traditional emulation and RTL modelling will not work. You need first spend an effort on architecture trade-off, not just in selecting components. This will mean that you will have to conduct traffic analysis, application partitioning, system sizing and impact of different types of physical layer. Also, depending on the application the benchmark will be very different.

The UCIe specification is new and there are no clear benchmarks. Also, the UCIe specification only provides guidance on latency and power. Both are stringent requirements. This means that a Power-Performance-Area study is evitable. As you have protocol-protocol-protocol conversion such as PCIe 6.0 to UCIe to AXI, the modelling setup is complex.

One solution is to look at system-modeling using VisualSim from Mirabilis Design. They have recently launched a UCIe System-level IP model and will be demonstrating a number of use cases of the interconnect at the Chiplet Summit. To guide designers along, they have published a guide with lots of use cases, expected power-performance results and options for optimization. They have both a paper presentation and a booth at the Summit. I hope to see you there!

Also, here is the link for a paper that folks can get: Performance modeling of a heterogeneous computing system based on the UCIe Interconnect Architecture

Abstract:

Today’s complex chip designs at leading-edge nodes generally consist of multiple dies (or chiplets). The approach allows for dies from different manufacturers or processes, as well as reusable IP. Designers need a system level model to evaluate different implementations of such complex situations.

An example system consists of an I/O chiplet, low power core chiplet, high-performance core chiplet, audio-video chiplet, and analog chiplet, interconnected using the Universal Chiplet Interconnect Express (UCIe) standard.

Our team considered several scenarios and configurations including advanced and standard packages, varied traffic profiles and resources, and a retimer to extend the reach and evaluate events on timeout. Identifying the strengths and weaknesses of the UCIe interconnect for mission applications helped us obtain the optimal configuration for each subsystem to meet performance, power, and functional requirements.

About Mirabilis Design Inc.

Mirabilis Design is a Silicon Valley software company, providing software and training solutions to identify and eliminate risk in the product specification, accurately predicting the human and time resources required to develop the product, and improve communication between diverse engineering
teams.

VisualSim Architect combines Intellectual Property, system-level modeling, simulation, environment analysis and application templates to significantly improve model construction, simulation, analysis and RTL verification. The environment enables designers to rapidly converge to a design which meets a diverse set of interdependent time and power requirements. It is used very early in the design process in parallel with (and as an aid to) the written specification and before an implementation (for example, RTL, software code, or schematic) of the product.

Also Read:

WEBINAR: How to Achieve 95%+ Accurate Power Measurement During Architecture Exploration

Mapping SysML to Hardware Architecture

Model-Based Design Courses for Students

Posted on

A Rare Offer from The SHD Group – A Complimentary Look at the RISC-V Market

A Rare Offer from The SHD Group – A Complimentary Look at the RISC-V Market
by Mike Gianfagna on 01-30-2024 at 10:00 am
Categories: AI, Andes Technology, IP, RISC-V

A Rare Offer from The SHD Group – A Complimentary Look at the RISC V Market

The web is a wonderful place to find information on almost any topic. While top-level information is easy to find, a deep dive often requires the services of a market research firm. These organizations specialize in “going deep” on many technology topics, offering insights not available with a Google search. And these services aren’t typically free. Access to focused research can get pricey. So, when a top-drawer research firm offers an 80-page report on a topic as hot as RISC-V for free, that gets my attention. Read on to learn about a rare offer from the SHD Group – a complimentary look at the RISC-V market.

About The SHD Group

The SHD Group was formed about four years ago. The organization brings together skills and services that focus on go-to-market strategies, marketing, market analysis, business development, sales pipeline building and closure on profitable opportunities. It operates across a range of markets, including AI, semiconductor, smart sensors, RISC-V, consumer and automotive.

The leadership team has decades of experience spanning all of these areas and more. The company offers first-hand executive experience, know-how and expertise in AI “from Edge to Cloud & back”.  You can learn more about the breadth and depth of experience offered by this team here. I want to take a moment to focus on the author of the RISC-V report.

Rich Wawrzyniak

Rich Wawrzyniak brings over 35 years of semiconductor industry experience to the organization. Previously, Rich spent 20 years focusing on market analysis at Semico Research. His demonstrated skills span sales management to corporate planning, with an expertise in ASICs, SoCs, SIP, memory, and design starts, as well as emerging areas like AI, RISC-V, and chiplets.

I personally know Rich from his time at Semico when I was in both ASIC and EDA. I have great memories of many in-depth conversations that lead to insightful analysis that helped the companies I worked for in measurable ways. Rich has an inquisitive and highly analytical nature – he gets to the facts that matter and presents results with compelling detail and accuracy. All this fueled my enthusiasm when I found out some of his work was being offered to all for free. This is a great opportunity.

About the RISC-V Report

I attended the recent RISC-V Summit and I can tell you the movement is gaining momentum fast. Open source has found its way into mainstream chip design. It’s exciting to watch. There are a lot of moving parts across many markets, so a comprehensive report that puts all this into perspective is quite valuable. The report offers a top-level, global market view, containing 80 pages, 30 tables, 16 figures and an accompanying ecosystem guide.

Some highlights covered in the report include:

RISC-V SoC Market Growth

  • RISC-V-based SoC unit shipments are forecast to surge to 16.2B units, with revenues reaching $92B by 2030, boasting CAGRs of 44% and 47%, respectively

SoC Market Growth

  • SoC architectures utilizing third-party IP exhibit substantial growth in units and revenues across industrial, automotive, networking, computer, consumer, and other categories, notably driven by the burgeoning AI market
  • Projections indicate SoC unit shipments reaching 69B units and revenues hitting $416B by 2030, showcasing CAGRs of 12.5% and 8.7%, respectively

SoC Design Starts

  • SoC design starts for SoCs using RISC-V CPU cores are forecast to reach 1,371 designs by 2030, a 15.7% CAGR
  • Design starts for consumer applications are expected to show the largest number of designs by 2030, with computer and networking applications following closely behind

Third-Party IP Market

  • In 2022, the worldwide IP market reached $7.9B, marking an 8.4% growth from 2021. Forecasts predict a 5.3% increase to $8.3B in 2023, projecting a potential $15B market by 2030, with a CAGR of 9%
  • The central processing unit (CPU) IP market soared by 22.4% in 2022 to $2.7B and is anticipated to hit $5.8B by 2030, demonstrating a robust 10.4% CAGR
  • RISC-V IP revenues surged to $156M in 2023, with an estimated CAGR of 39.5% through 2030

Comments from the Author

 I had the opportunity to chat with the report’s author, Rich Wawrzyniak recently. I was looking for a more color about how the report was developed and some candid comments from Rich regarding what he discovered. First, Rich explained this was a large project. He surveyed and spoke with about 32 companies spanning IP vendors, software companies, EDA companies, device manufacturers, and end users.  A huge amount of data was compiled as a result of this exercise.

I asked Rich if there were any surprises in the results. He mentioned one regarding classification in the IP segment. It turns out the third largest segment here is do-it-yourself. That is, those who weren’t using a specific RISC-V vendor but rather building their own design. While this type of behavior doesn’t scale well, it highlights the substantial amount of exploration that is going on in the market today around RISC-V.

And that fact illuminates some of the unique attributes of the emerging RISC-V market. We spent some time discussing the impact AI has had here. Rich explained that the ability to customize the ISA extensions RISC-V offers SoC designers allows for fine-tuning of the ISA to more fully support the fast-paced AI algorithm developments that are now occurring in the market.

He went on to say that using this capability, SoC designers can more closely tailor their silicon solutions to match changing market requirements. This is especially important when they are considering development of domain-specific solutions for a wide range of applications. How that accelerator is built will have a dramatic impact on the success of any new AI technology. It’s all about speed and power efficiency. And every new idea has its own unique set of success factors.

The extensible nature of the RISC-V architecture and the rich community of innovators fuel the impact this movement is having on product development. A perfect storm of innovation supply and demand if you will. Rich felt this market is just on the cusp of explosive growth. Where all this takes us is hard to predict, but watching the progress will be quite exciting.

RISC-V International encouraged The SHD Group to launch this analysis effort and did encourage its membership to speak with the organization. The full version is likely the most comprehensive analysis of the RISC-V market available today. Rich has a long history of developing world-class market research – this project represents some outstanding and important work. 

To Learn More

You can download your complimentary copy of the 80-page report here.

The extensive full report spanning over 225 pages with 107 tables and 89 figures is also available for purchase. This report provides detailed insights into the current and future projections of the RISC-V market up to 2030. It covers most aspects of the RISC-V market, including end applications, device types, design starts, IP and global projections by region. The report is intended to provide a valuable analysis for business strategists, investors, and technology companies that require deep analysis and granular data. To purchase the full report, email info@theshdgroup.com.

And those are the details about a rare offer from the SHD Group – a complimentary look at the RISC-V market.

Posted on

KLAC- OK Quarter & flat guide- Hopefully 2025 recovery- Big China % & Backlog

KLAC- OK Quarter & flat guide- Hopefully 2025 recovery- Big China % & Backlog
by Robert Maire on 01-30-2024 at 6:00 am
Categories: Semiconductor Advisors, Semiconductor Services

KLAC Foundry Logic

– KLAC reported an OK QTR & flat guide-waiting for 2025 recovery?
– China exposure remains both risk & savior & big in backlog
– Wafer inspect strong- Patterning on long slide- PCB biz for sale
– Some bright spots but memory still weak- Foundry/Logic OK

Bumping along the bottom of the cycle looking towards a hopeful 2025

KLAC reported $2.49B in revenues and $6.16 in non GAAP EPS. The street was at $2.46B in revenues and EPS of $5.91, so a little better as usual.

Guidance was more or less flattish to slightly down at $2.3B +-$125M and EPS of $5.26+- $0.60.

One of the reasons for March being down is some business that is slipping out of the quarter into future quarters making March lower and future quarters higher. So while technically March is a bottom, its NOT because its the bottom of the cycle but rather customer orders shifting around. Other than that we heard the same story that we heard from Lam that the back half of the year should be better than the front half (we have heard that before)

China is also “stable” at 41% of business but still a risk

China remains the double edge sword of both a risk and savior at the same time. Without the inflated China business KLA would be in a world of hurt (along with other equipment makers) but we continue to wonder how long its going to last.

Its very interesting to note that China is clearly making up a huge portion of the companies backlog as new Chinese buyers are putting down deposits to secure a place in line hopefully before any sanctions kick in. Perhaps the view is that by having an order in they will eventually get a tool.

We obviously see risk in this and remain concerned about the huge amount of business China represents on both a quarterly basis as well as in backlog.

Selling off the PCB business?

Management announced they were looking at strategic alternatives for the PCB business they acquired as it represents less than 1% and has not been great for KLA.

Our observation is that the acquisition of Orbotech almost 6 years ago has clearly not worked out as well as expected .

In our view, the $3.4B acquisition price would likely have been better spent inside KLA’s wheelhouse rather than venturing outside to try to diversify. While the aim to diversify was a good one, especially when comparing KLA to more diversified equipment companies like Applied Materials, the results of the effort have been less than stellar.

Unfortunately given todays political environment there is not a lot in KLA’s wheelhouse that they could buy so it is likely best to double down on existing markets and inside the company.

Wafer inspection strong while Patterning continues long slide

We would note that wafer inspection has become the lions share of KLA’s business at 47% while pattering was about a third of that coming in at 17% of overall revenues. Patterning was down 50% year over year and 21% quarter over quarter while wafer inspection was down a very small 7% year over year and up 15% quarter over quarter. For the year in total, wafer inspection was only off 5% while patterning was off by four times that at 20%.

While these businesses are lumpy from one quarter to the next and account for a lot of variations, the long term pattern is quite clear that KLA’s dominance in patterning is in decline and the growth is slowing significantly especially as compared to wafer inspection. This is obviously a significant change from the long term model where both markets were the two pillars of KLA’s business.

DRAM is somewhat alive with NAND still dead

DRAM represented 85% of memory business while NAND was at near zero levels of 15%, not much different than we have heard from other equipment makers.

HBM memory and DDR5 are bright spots that are driving DRAM while there remains a ton of excess and unutilized capacity in the market that we expect will take at least a year or more to sop up so we don’t expect a major broad recovery any time soon but specific area strength in memory

The Stocks

KLAC was down about 5% in the after market. Obviously the bad guide out of Intel did not help a luke warm report with no definitive recovery other than sometime in 2025.

We did not hear much different out of KLA than we heard from Lam and would expect similar stock performance from both as well as AMAT.

It is clear that a real recovery is a year away and memory will be slow to recover which will keep the pace of recovery in check.

China remains a both risk and reward for all three companies at 40% ish of business.

The stocks continue to trade at relatively high multiples for companies that are still in a downcycle. There is still a lot that can happen before we get to a real recovery and we don’t even know the slope or speed of the recovery other than a hope that it will be in 2025.

About Semiconductor Advisors LLC

Semiconductor Advisors is an RIA (a Registered Investment Advisor),
specializing in technology companies with particular emphasis on semiconductor and semiconductor equipment companies. We have been covering the space longer and been involved with more transactions than any other financial professional in the space. We provide research, consulting and advisory services on strategic and financial matters to both industry participants as well as investors. We offer expert, intelligent, balanced research and advice. Our opinions are very direct and honest and offer an unbiased view as compared to other sources.

Also Read:

ASML – Strong order start on long road to 2025 recovery – 24 flat vs 23 – EUV shines

2024 Semiconductor Cycle Outlook – The Shape of Things to Come – Where we Stand

Is Intel cornering the market in ASML High NA tools? Not repeating EUV mistake

Posted on

2024 Outlook with Steve Roddy of Quadric

2024 Outlook with Steve Roddy of Quadric
by Daniel Nenni on 01-29-2024 at 10:00 am
Categories: IP, Quadric

Man on Llama

Quadric Inc. is the leading licensor of general-purpose neural processor IP (GPNPU) that runs both machine learning inference workloads and classic DSP and control algorithms.  Quadric’s unified hardware and software architecture is optimized for on-device ML inference. I have know Steve Roddy for many years, he is a high standard in the IP business.

Tell us a little bit about yourself and your company.
Quadric is a startup processor IP licensing company delivering a unique general-purpose, programmable neural processor (GPNPU) IP solution. In a marketplace with more than a dozen machine learning “accelerators” ours is the only NPU solution that is fully C++ programmable that can run any and every AI/ML graph without the need for any fallback to a host CPU or DSP. With more than 25 years of marketing and management experience in the IP business, I lead the marketing and product management teams at Quadric.

What was the most exciting high point of 2023 for your company?
2023 was exciting for Quadric because it marked the debut of our first licensable IP product in May 2023 – both first production RTL deliveries of the Chimera GPNPU and the launch of our online Quadric DevStudio. In the seven months since we’ve been expanding or our sales & FAE team around the world. 2023 was an eventful and successful year, indeed.

What was the biggest challenge your company faced in 2023?
The biggest “news” in 2023 in the market for NPUs/GPNPUs was the dramatic upsurge in interest in Large Language Models (LLMs) in devices, rather than running purely in the cloud. Whether it is the rise of the so-called “AI PC” or the embedded of LLM-based voice assistants in countless end products, the surge in user demand for transformer-based LLMs dramatically impacted the NPU IP market. Many existing NPUs could not efficiently run LLMs, putting stress on the silicon ecosystem.

How is your company’s work addressing this biggest challenge?
Unique among NPU vendor offerings, the Quadric Chimera GPNPU is 100% programmable. As a result, we tackled the on-device LLM wave by demonstrating the Llama-2 LLM less than 5 weeks after it was published. Meanwhile our competitors were announcing new cores that won’t be available until mid-2024 (or later). The rate of change of LLMs has only accelerated since mid-2023 with a myriad of new language model type and topologies. This rapid pace of change demands flexible hardware that can run as-yet not invented machine learning models.

What do you think the biggest growth area for 2024 will be, and why?
The ML inference market that we serve is rapidly changing. The dominant ML algorithm styles of three and four years ago were classic convolution-based CNNs, such as the Resnet and MobileNet and Yolo families of networks. Today, newer structures leveraging transformer topologies – such as LLMs and ViT models – are rapidly displacing the older CNNs. That is turn is causing silicon design teams to respin older devices to be positioned to support these newer algorithms in the coming years.

How is your company’s work addressing this growth?
Quadric is continuously adding ports of new algorithms to our processors. Adding demonstration of a new ML model is a pure software effort for us, and we are focused in 2024 on widening the array of models further with each periodic software release. Today we support all the major modalities of ML inference, including a variety of leading-edge transformers.

What conferences did you attend in 2023 and how was the traffic?
In 2023 we attended smaller, focused technical conferences (Embedded Vision Summit, DAC, Design Solution Forum). We avoided the bigger mass-gathering shows in 2023 (CES, MWC, Embedded World) because those shows were not all the way “back” to pre-pandemic attendance levels. However, I did go to CES 2024 this month for some very interesting meetings and to take a pulse of the marketplace.

Will you attend conferences in 2024? Same or more?
I think the world has fully returned to “normal” in 2024. CES in Las Vegas this month was a good indicator – full crowds reminiscent of 2019. As a result we will be attending both the small, focused IP-centric conferences this year as well as the broader shows.

Also Read:

Fast Path to Baby Llama BringUp at the Edge

Vision Transformers Challenge Accelerator Architectures

An SDK for an Advanced AI Engine

Posted on

LRCX- In line Q4 & flat guide- No recovery yet- China still 40%- Lags Litho

LRCX- In line Q4 & flat guide- No recovery yet- China still 40%- Lags Litho
by Robert Maire on 01-29-2024 at 6:00 am
Categories: Semiconductor Advisors, Semiconductor Services
1 Comment

Lam Research LCRX

– Lam reported as expected and guided flat- No recovery yet
– Some mix shifts but China still 40% (8X US at 5%)-NVM still low
– HBM is promising but Lam needs a broad memory recovery
– Lam has not seen order surge ASML saw- Likely lagging by 3-4 QTRs

An in line quarter and uninspiring flat guide for Q1

As compared to ASML’s huge order report this morning Lam put up relatively in line numbers and a flat guide with no visibility on an upturn. Revenues were $3.76B and EPS of $7.52 with the year at $14.3B and $27.33EPS.

Although the company suggested that overall WFE would go from low $80B 2023 to mid $80B in 2024 we did not hear that they were on the road to recovery just yet.

The flat guide coupled with conservative language makes it clear that we are not yet in recovery mode. This sense was underscored by a headcount reduction and keeping expenses and inventory under control.

They talked about memory fab utilization remaining low DRAM was 31% of business , NAND remained relatively low at 17%. Service was $1.46B or about 39% of business so tools sales continue to be very weak.

China is “stable” at 40% of business- Could represent over 50% of tool sales

Lam still remains highly dependent upon China which was 40% of business and according to management will likely remain at elevated levels. This compares to the US at 5%, so China is outspending the US at Lam by eight to one.

We think this exposure remains an overall negative on the story. If we back out the service business which is obviously dominant in older markets we could imagine that China is likely over 50% of Lam’s sales which is exposed if there are any serious restrictions imposed.

Lam will likely lag ASML by 3-4 quarters

given the lead times of litho tools versus the turns business that are Lam’s tools we would expect Lam to see an order pick up in the second half or end of 2024. Customers are not going to buy dep and etch tools without having a litho tool to drive patterning.

This would tend to imply a flattish 2024 overall with perhaps some pick up at the end of the year.

High Bandwidth memory and DDR5 are bright spots

As we have mentioned many times HBM will clearly be great given how strongly it will be driven by AI growth, however HBM is a relatively small part of the overall memory market. We also remain concerned about potential oversupply and price collapse as all HBM makers are rushing to put on more capacity or move existing other DRAM capacity to HBM. There will clearly be a bunch of HBM spend in the near term but Lam needs a much broader recovery in the broader memory market.

The NAND/NVM market remains oversupplied with capex spending at historical lows.

With the current and continued oversupply in memory we don’t see the need for any new memory fabs in quite some time. What we will likely see are mainly upgrades in existing fabs to HBM or DDR5.

Lam has historically been the poster child for memory and Korea related to it. Lam has done a good job in the weak memory market by diversifying into foundry/logic. China has obviously made up for much of the Korean weakness as well.

The Stocks

Lam was flat in the after market after being up 2% on the strong ASML news.

We are not motivated to go out and buy either Lam or AMAT based on Lam’s mediocre earnings call. There is not enough evidence of a near term recovery and the stock is already trading at a premium that we believe is above what the stock deserves given where the company is in the cycle.

Given the differential between what we heard from ASML and Lam, ASML remains our clear preference.

We do think Lam will obviously recover but after ASML and the recovery may not be nearly as strong as ASML which is driven by EUV and the High NA introduction.

We have maintained the view that the memory over supply is large. If we are just now starting to see an uptick in memory fab utilization it will likely be a much longer time before they start buying more NAND equipment and even the bright spots of HBM and DDR5 are not nearly enough to make up the difference.

About Semiconductor Advisors LLC

Semiconductor Advisors is an RIA (a Registered Investment Advisor), specializing in technology companies with particular emphasis on semiconductor and semiconductor equipment companies. We have been covering the space longer and been involved with more transactions than any other financial professional in the space. We provide research, consulting and advisory services on strategic and financial matters to both industry participants as well as investors. We offer expert, intelligent, balanced research and advice. Our opinions are very direct and honest and offer an unbiased view as compared to other sources.

Also Read:

2024 Semiconductor Cycle Outlook – The Shape of Things to Come – Where we Stand

Is Intel cornering the market in ASML High NA tools? Not repeating EUV mistake

Podcast EP205: A Multi-Decade View of Process and Device Innovation at Intel with Paul Fischer

Podcast EP205: A Multi-Decade View of Process and Device Innovation at Intel with Paul Fischer
by Daniel Nenni on 01-26-2024 at 10:00 am
2 Comments

Dan is joined by Paul Fischer. Paul is the director of Chip Mesoscale Processing in Intel’s Components Research. He and his team are currently working on Gallium Nitride for energy efficient power delivery and RF communications, and technologies for heterogeneous monolithic integration.

Paul discusses the innovations he’s seen during his 30-year career at Intel, starting with the 130 nm process node. Paul describes the breadth and depth of material, device and process developments at Intel during his time there.

He discusses some of the advances presented by Intel at the recent IEDM across backside power delivery, vertical transistor stacking, the energy efficiency offered by gallium nitride devices and 2D material innovations. The combination of this work will open new pathways for continued Moore’s Law scaling of devices and systems.

The views, thoughts, and opinions expressed in these podcasts belong solely to the speaker, and not to the speaker’s employer, organization, committee or any other group or individual.

Posted on

ASML – Strong order start on long road to 2025 recovery – 24 flat vs 23 – EUV shines

ASML – Strong order start on long road to 2025 recovery – 24 flat vs 23 – EUV shines
by Robert Maire on 01-26-2024 at 6:00 am
Categories: Semiconductor Advisors, Semiconductor Services
3 Comments

ASML Cleanroom EUV Wafer Stage Training

– ASML orders more than triple sequentially- Utilization increases
– Management remains conservative with flat revenues 2024 vs 2023
– Recovery will be slow, targeting 2025- Long & weak cyclical bottom
– Litho orders are leading indicator of future wider recovery

Strong orders pave the way for a transition 2024 & recovery in 2025

ASML reported revenues of Euro 7.2B and EPS of Euro 5.21 with gross margins of 51.4%. Outlook is between Euro 5B and 5.5B in Q1 2024 for a variety of factors as the industry moves through the bottom of the cycle.

2023 was a great year for ASML with business up 30% year over year.

We continue to view China restrictions as essentially non impactful as other business clearly makes up for it.

Most importantly orders more than tripled quarter over quarter to Euro 9.2B a new record. Given lead times on systems, most of these orders will contribute to revenue in 2025 rather than 2024. Weakness in Q1 is evidence of prior weak orders and continued slow industry environment.

Perhaps just as important as the huge order number is the fact that we are now up to Euro 39B in backlog. This huge backlog is key to future financial and management performance as it represents almost a year and a half of future business. This helps improve business planning and more importantly manage the cyclicality of the business.

Given the large price tag of EUV systems and especially High NA systems, we would almost focus more on backlog as orders will be lumpy given the large numbers involved

Utilization increasing is a positive sign for the industry and future business

ASML can monitor the pulse of the industry on a daily basis by looking at the utilization of its tools which clearly has been down over the down cycle. ASML is now seeing an up tick in utilization after a long decline which is a good indicator that we are transitioning through the bottom of the cycle.

After a strong logic/foundry business in 2023 there is an expectation that there will be some “digestion” of all those sales coupled with signs of life coming out of the memory business, so we will see a shift in market share towards memory.

Conservative outlook for 2024 flat with 2023

Management suggested that the second half of 2024 will be better than the first half which we would expect in a “transition year” to a recovery in 2025.

We would remind investors that the quarters strength in orders won’t become revenue until 2025 or late 2024 at best.

This one year or longer lag effect between orders and revenue obvioulsy explains the weak Q1 revenue expectations but obviously investors will be able to look through this

In line with our earlier note

We put out a note two days ago that previewed 2024 and we suggested that 2024 would be a slow recovery but a recovery none the less. Our views were echoed by the conservative outlook for ASML in 2024. We agree that 2024 will be a “transition” year. We would also suggest that lithography will lead the industry out of the downturn as litho systems are always the first ordered given their long lead times, we would expect metrology and inspection to follow after which will come dep and etch and other process tools.

We pointed out ASML as our top pick in the industry and we were not disappointed. ASML continues to dominate the equipment industry as the only true monopoly in town. ASML will likely continue to take capex and semiconductor equipment market share away from others as the overall spend on litho accelerates especially given the cost of EUV and specifically High NA. One high NA tool could easily be the cost of 50 process tools. made by others in the industry.

China is a non issue

We maintain our view that the restrictions on sales to China is not a significant issue in light of the huge order numbers without China immersion tools. Being able to sell into China would be a few Euros more but is less significant when you are capacity constrained.

We also continue to point out many times that the semiconductor industry is a zero sum game…..chips not made in China due to technology restrictions will be made elsewhere and ASML will ship those advanced litho tools to those elsewhere places where there are no restrictions. If those places happen to be new fabs in the US or Europe or Japan, so much the better.

At this point, even artificial restrictions which shift chip capacity away from China who has been trying to spend their way into market dominance, would be a good thing as the world needs to disperse chip capacity around the globe and not watch China take over yet another industry through market force.

The Stocks

The 10% reaction in ASML’s stock was even better than we expected and took much of the semiconductor stock market along with it.

We would point out that some process companies are likely trading at too high multiples as they do not have the ASML monopoly nor the share gains of ASML and will be late to the party as compared to ASML.

As we suggested in our last note we would remain more selective and focus on leaders and differentiated stories such as ASML & TSMC etc;.

We would also remind investors, as we did in the note, that its not just “off to the races”, that this is going to be a recovery that will take time and likely not gel until 2025.

There is a lot positive in the market such as AI and associated High Bandwidth Memory but we don’t have a very broad based macro recovery to drive every corner of the chip market just yet.

ASML did point out that foundry/logic will be weaker in 2024 versus memory which has been virtually dead for over 2 years. We would be cognizant of that in trying to be more selective in our stock picks as well.

But in the end its nice to know that things are finally getting better…..

About Semiconductor Advisors LLC

Semiconductor Advisors is an RIA (a Registered Investment Advisor), specializing in technology companies with particular emphasis on semiconductor and semiconductor equipment companies. We have been covering the space longer and been involved with more transactions than any other financial professional in the space. We provide research, consulting and advisory services on strategic and financial matters to both industry participants as well as investors. We offer expert, intelligent, balanced research and advice. Our opinions are very direct and honest and offer an unbiased view as compared to other sources.

Also Read:

2024 Semiconductor Cycle Outlook – The Shape of Things to Come – Where we Stand

Is Intel cornering the market in ASML High NA tools? Not repeating EUV mistake

AMAT- Facing Criminal Charges for China Exports – Overshadows OK Quarter

Posted on

Why Did Synopsys Really Acquire Ansys?

Why Did Synopsys Really Acquire Ansys?
by Daniel Nenni on 01-25-2024 at 10:00 am
Categories: Ansys, Inc., EDA, Synopsys
3 Comments

Synopsys Ansys Logos

Mergers and acquisitions have been a big part of EDA since the beginning. We keep an EDA/IP Mergers and Acquisitions Wiki, it is 13 years old now and has more than one million views. Personally, I have been involved with dozens of acquisitions over my 40 year career, some good, some bad, all are interesting and are an important part of EDA history.

If you have been listening to my quarterly EDA revenue podcasts with Wally Rhines you will know that EDA has been killing it lately with double digit revenue growth. Here is the thing, EDA revenue reports are from the big EDA companies while dozens of smaller EDA companies go unreported. It is those  EDA companies that have been acquired at an alarming rate over the past 10 years adding to the big EDA company’s growth numbers. Looking at the Wiki you can see that hundreds of mergers is what made EDA what it is today, three big dogs eating out of the same bowl, not a pretty site (Joe Costello quote).

I think the Siemens acquisition of Mentor Graphics was one of the all-time greats of course. Siemens has acquired quite a few EDA companies since then and will continue to do so, my opinion. From what I have heard the Siemens acquisition of Solido Design Automation (I worked for Solido) is viewed as one of the most successful Siemens EDA acquisitions thus far, to which I agree completely. It was a 1+1=10 type of deal. Siemens also acquired Fractal Technology (I worked for Fractal) which fits perfectly with the Siemens Solido group.

Prior to Siemens I worked for Tanner EDA (acquired by Mentor) and Berkeley Design Automation (acquired by Mentor) so I know them well. One of the most interesting acquisitions I worked on was S2C. A company out of Hong Kong seriously outbid the usual EDA suspects, it was quite an international experience.

So that is what I do during the day, I help small EDA/IP companies with their exit plan which is much more complex than it sounds and my recipe is secret, like the formula to Coke and the Kentucky Fried Chicken secret herbs and spices.

Quite a bit has been written about the Synopsys / Ansys acquisition so I will try and not be repetitive. I also asked Chat GPT about it and the response was absolutely ridiculous.

From my perspective the acquisition has been in process for a while. Synopsys and Ansys have been close partners for some time. There is no real overlap in products and the two companies are quite compatible. So, you have to ask about timing since Synopsys just finished its big CEO transition. The simple answer is that Synopsys was not the first bidder.

In my M&A experience getting the first term sheet is always the hardest and Synopsys is not the quickest to a term sheet. To get maximum value for your EDA company you must have multiple offers. The most I have personally seen is four offers but I usually see three or two offers or sometimes it is a CEO to CEO single offer type of deal.

Considering the top 3 EDA companies, the ones who could pay a premium price for Ansys, my guess would be that Cadence was the first bidder. Siemens has too much overlap with Ansys and there would be anti-trust concerns. The Ansys ($2B in 2022 revenue) acquisition would put Cadence ($3.5B 2022) ahead of Synopsys ($5B 2022) which is a very big deal given the competitive history of the two companies. Synopsys may also be getting out of the software integrity business which would be a revenue hole that must be filled. The Synopsys Ansys combined revenue will be close to $8B when 2023 is reported.

Of course there are many other reasons for Synopsys to acquire Ansys but, if I had to pick one, in my experience ego plays a very big role in M&A and Synopsys has the #1 ego in EDA and it is well deserved, absolutely.

Looking forward maybe Siemens EDA will acquire Cadence? That is the only hope if either company wants to catch up to Synopsys. Then there would be two big dogs eating out of the same bowl, not a pretty sight.

Congratulations Synopsys and Ansys! Well played!

Also Read:

Synopsys Geared for Next Era’s Opportunity and Growth